未验证 提交 d3366853 编写于 作者: Y ykkk2333 提交者: GitHub

migrate squeeze kernel to phi, test=kunlun (#45968)

上级 e6d397e6
/* Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/fluid/operators/squeeze_op.h"
#ifdef PADDLE_WITH_XPU
namespace ops = paddle::operators;
namespace plat = paddle::platform;
REGISTER_OP_XPU_KERNEL(
squeeze,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, float>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, double>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, plat::float16>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, bool>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, int>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, uint8_t>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, int8_t>,
ops::SqueezeKernel<paddle::platform::XPUDeviceContext, int64_t>);
REGISTER_OP_XPU_KERNEL(
squeeze_grad,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, float>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, double>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, plat::float16>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, bool>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, int>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, uint8_t>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, int8_t>,
ops::SqueezeGradKernel<paddle::platform::XPUDeviceContext, int64_t>);
REGISTER_OP_XPU_KERNEL(
squeeze2,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, float>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, double>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, plat::float16>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, bool>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, int>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, int8_t>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, uint8_t>,
ops::Squeeze2Kernel<paddle::platform::XPUDeviceContext, int64_t>);
REGISTER_OP_XPU_KERNEL(
squeeze2_grad,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, float>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, double>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, plat::float16>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, bool>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, int>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, int8_t>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, uint8_t>,
ops::Squeeze2GradKernel<paddle::platform::XPUDeviceContext, int64_t>);
#endif
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/phi/kernels/squeeze_grad_kernel.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/impl/squeeze_grad_kernel_impl.h"
PD_REGISTER_KERNEL(squeeze_grad,
CPU,
ALL_LAYOUT,
phi::SqueezeGradKernel,
float,
double,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/phi/kernels/squeeze_kernel.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/impl/squeeze_kernel_impl.h"
PD_REGISTER_KERNEL(squeeze,
CPU,
ALL_LAYOUT,
phi::SqueezeKernel,
float,
double,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
PD_REGISTER_KERNEL(squeeze_with_xshape,
CPU,
ALL_LAYOUT,
phi::SqueezeWithXShapeKernel,
float,
double,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/phi/kernels/squeeze_grad_kernel.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/impl/squeeze_grad_kernel_impl.h"
PD_REGISTER_KERNEL(squeeze_grad,
GPU,
ALL_LAYOUT,
phi::SqueezeGradKernel,
float,
double,
phi::dtype::bfloat16,
phi::dtype::float16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include "paddle/phi/common/int_array.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/core/tensor_utils.h"
namespace phi {
template <typename T, typename Context>
void SqueezeGradKernel(const Context& dev_ctx,
const DenseTensor& xshape,
const DenseTensor& dout,
const IntArray& axes,
DenseTensor* dx) {
auto xshape_dims = xshape.dims();
auto x_dims = phi::slice_ddim(xshape_dims, 1, xshape_dims.size());
dev_ctx.template Alloc<T>(dx);
phi::Copy(dev_ctx, dout, dev_ctx.GetPlace(), false, dx);
dx->Resize(x_dims);
}
} // namespace phi
...@@ -12,20 +12,35 @@ ...@@ -12,20 +12,35 @@
// See the License for the specific language governing permissions and // See the License for the specific language governing permissions and
// limitations under the License. // limitations under the License.
#include "paddle/phi/kernels/squeeze_kernel.h" #include "paddle/phi/kernels/squeeze_grad_kernel.h"
#include "paddle/phi/backends/gpu/gpu_context.h" #include "paddle/phi/backends/all_context.h"
#include "paddle/phi/core/kernel_registry.h" #include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/impl/squeeze_kernel_impl.h" #include "paddle/phi/core/tensor_utils.h"
PD_REGISTER_KERNEL(squeeze, namespace phi {
GPU, template <typename T, typename Context>
void SqueezeGradKernel(const Context& dev_ctx,
const DenseTensor& xshape,
const DenseTensor& dout,
const IntArray& axes,
DenseTensor* dx) {
auto xshape_dims = xshape.dims();
auto x_dims = phi::slice_ddim(xshape_dims, 1, xshape_dims.size());
dev_ctx.template Alloc<T>(dx);
phi::Copy(dev_ctx, dout, dev_ctx.GetPlace(), false, dx);
dx->Resize(x_dims);
}
} // namespace phi
PD_REGISTER_KERNEL(squeeze_grad,
CPU,
ALL_LAYOUT, ALL_LAYOUT,
phi::SqueezeKernel, phi::SqueezeGradKernel,
float, float,
double, double,
phi::dtype::bfloat16, phi::dtype::bfloat16,
phi::dtype::float16,
bool, bool,
int, int,
uint8_t, uint8_t,
...@@ -34,14 +49,15 @@ PD_REGISTER_KERNEL(squeeze, ...@@ -34,14 +49,15 @@ PD_REGISTER_KERNEL(squeeze,
phi::dtype::complex<float>, phi::dtype::complex<float>,
phi::dtype::complex<double>) {} phi::dtype::complex<double>) {}
PD_REGISTER_KERNEL(squeeze_with_xshape, #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
PD_REGISTER_KERNEL(squeeze_grad,
GPU, GPU,
ALL_LAYOUT, ALL_LAYOUT,
phi::SqueezeWithXShapeKernel, phi::SqueezeGradKernel,
float, float,
double, double,
phi::dtype::bfloat16,
phi::dtype::float16, phi::dtype::float16,
phi::dtype::bfloat16,
bool, bool,
int, int,
uint8_t, uint8_t,
...@@ -49,3 +65,21 @@ PD_REGISTER_KERNEL(squeeze_with_xshape, ...@@ -49,3 +65,21 @@ PD_REGISTER_KERNEL(squeeze_with_xshape,
int64_t, int64_t,
phi::dtype::complex<float>, phi::dtype::complex<float>,
phi::dtype::complex<double>) {} phi::dtype::complex<double>) {}
#endif
#ifdef PADDLE_WITH_XPU
PD_REGISTER_KERNEL(squeeze_grad,
XPU,
ALL_LAYOUT,
phi::SqueezeGradKernel,
float,
double,
phi::dtype::float16,
bool,
int,
uint8_t,
int8_t,
int64_t) {}
#endif
...@@ -12,8 +12,10 @@ ...@@ -12,8 +12,10 @@
// See the License for the specific language governing permissions and // See the License for the specific language governing permissions and
// limitations under the License. // limitations under the License.
#pragma once #include "paddle/phi/kernels/squeeze_kernel.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/backends/all_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/core/tensor_utils.h" #include "paddle/phi/core/tensor_utils.h"
#include "paddle/phi/kernels/funcs/unsqueeze.h" #include "paddle/phi/kernels/funcs/unsqueeze.h"
...@@ -43,3 +45,94 @@ void SqueezeWithXShapeKernel(const Context& dev_ctx, ...@@ -43,3 +45,94 @@ void SqueezeWithXShapeKernel(const Context& dev_ctx,
} }
} // namespace phi } // namespace phi
PD_REGISTER_KERNEL(squeeze,
CPU,
ALL_LAYOUT,
phi::SqueezeKernel,
float,
double,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
PD_REGISTER_KERNEL(squeeze_with_xshape,
CPU,
ALL_LAYOUT,
phi::SqueezeWithXShapeKernel,
float,
double,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
PD_REGISTER_KERNEL(squeeze,
GPU,
ALL_LAYOUT,
phi::SqueezeKernel,
float,
double,
phi::dtype::float16,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
PD_REGISTER_KERNEL(squeeze_with_xshape,
GPU,
ALL_LAYOUT,
phi::SqueezeWithXShapeKernel,
float,
double,
phi::dtype::float16,
phi::dtype::bfloat16,
bool,
int,
uint8_t,
int8_t,
int64_t,
phi::dtype::complex<float>,
phi::dtype::complex<double>) {}
#endif
#ifdef PADDLE_WITH_XPU
PD_REGISTER_KERNEL(squeeze,
XPU,
ALL_LAYOUT,
phi::SqueezeKernel,
float,
double,
phi::dtype::float16,
bool,
int,
uint8_t,
int8_t,
int64_t) {}
PD_REGISTER_KERNEL(squeeze_with_xshape,
XPU,
ALL_LAYOUT,
phi::SqueezeWithXShapeKernel,
float,
double,
phi::dtype::float16,
bool,
int,
uint8_t,
int8_t,
int64_t) {}
#endif
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册